After two years of data acquisition and processing, we are proud to release the NUS 3D point cloud dataset.
The NUS panoramic 3D provides a large-scale 3D point cloud dataset of NUS campus and a comprehensive learning benchmark for visual recognition, scene understanding and varies kinds of vision problems. There are over 1,000,000,000 points from 5 km^2 area in dataset, with the rich information of global XYZ, RGB value, annotations of classification and instance.
In more details, point cloud of NUS campus is divided into six regions: FOE, FASS, Utown, PGP, UCC and Ridge area based on their locations, functions and architectural features. All points in six regions are in the same coordinate reference system. We also present statistics and distribution of each region as an interactive map in Description. Owe to the varies appearance and functionalities of regions, we deliberately assign each region into training set or testing set when establishing the benchmark, to ensure that two sets are independently sampled with less similarity, which is shown in Data.Our dataset annotates the point cloud into 14 categories and 3000 instances, which are presented in the format of tree graph in Description.For the convenience of dataset referrers, we provide tools format transformation and label modification, which allows the referrers to flexibly merge categories for different tasks.
To the best of our knowledge, this dataset is the largest aerial laser scanning (ALS) 3D point cloud data with instance level semantic and geometric annotation. We are aiming at benchmarking learning methods to various ALS 3D problems such as (but not limited to): segmentation, detection as well as classification. This will make contributions for research progress in various areas like computer vision, geographic information system, computer graphics and drive new solutions and problems in related areas such as 3D model reconstruction, urban planning, autonomous driving and etc.